Benjamin Perry

Designing enzymes with bots and bits! | Romero Lab at Duke

[Benjamin Perry @bots_and_bits]

New Brand Name Idea: ThermoFishy🐟

[nature @Nature]

Catalogue entries for more than 100 antibodies sold by the research services and supply company Thermo Fisher Scientific contain images that have apparently been manipulated, according to a pair of science sleuths.

https://t.co/5rUUgxceoE

[Benjamin Perry @bots_and_bits]

@sokrypton for protein design around a ligand you could even reverse it to masking out the dog instead. brilliant.

[Benjamin Perry @bots_and_bits]

@joe_fenrir @sokrypton Was wondering the same thing and discovered they have it on HuggingFace here: https://t.co/mCHvOQVPWj

haven't tested it yet

[Benjamin Perry @bots_and_bits]

@addictedtoigno1 Thermo retraction incoming?👀

[Benjamin Perry @bots_and_bits]

Experimentally determined protein folding stability dataset on metagenomic sequences + predictive models trained on it.

Fantastic work from @ChoYehlin, @KotaroTsuboyama, and others🧑‍🔬

[Yehlin Cho @ChoYehlin]

🚀 Excited to share our new work: Absolute Stability Predictor!

📊: https://t.co/gtgQjPRAX6

Built the MGnify Stability Dataset (1.8M+ measurements) and developed stability prediction models, together with @grocklin, @KotaroTsuboyama, @sokrypton, and teams.

[Benjamin Perry @bots_and_bits]

@bffswithbiology that's not an american enough unit. have you tried representing your volumes in teaspoons, drops, or # of coca-cola cans?

[Mohamed Sabba @ma_sabba]

The backlash to this perfectly reasonable rule, largely from the usual suspects (i.e. AI grifters) is appalling.

I hate to break it to you but arXiv is used by actual professional scholars who are not going to throw away basic academic standards for the sake of absolute garbage.

[Benjamin Perry @bots_and_bits]

this is a great move from arxiv

[Andrew White 🐦‍⬛ @andrewwhite01]

hallucinated references will land you a 1-year ban from arxiv now. wow

[Benjamin Perry @bots_and_bits]

Presenting at the GEM bio workshop poster session at 11:05 AM!

Room 210📍

[Benjamin Perry @bots_and_bits]

I'll be in Rio this week for #ICLR2026 🇧🇷 including presenting AlphaFast and LIghtning-Boltz at the GEM workshop (4/27 in room 210)

DMs are open to chat about all things accelerating bio-molecular structure prediction and contrastive learning for protein design!🧬

[Benjamin Perry @bots_and_bits]

Wonder if they are using flow matching or diffusion?

[Stinson Dean 🌲🪓 @LumberTrading]

This is a baseball bat

[Benjamin Perry @bots_and_bits]

@ChrisHayduk @karpathy you are totally right pedagogical materials in AI x Bio are sorely needed. fantastic work Chris!

[Chris Hayduk @ChrisHayduk]

A couple of months ago, I announced that I was partway through implementing a simple, readable AlphaFold2 in pure PyTorch, inspired by @karpathy's minGPT.

Today, I'm happy to share minAlphaFold2 - the completion of that project.

Repo link: https://t.co/bU59VUm5sB

[Benjamin Perry @bots_and_bits]

A workshop with many fantastic organizers! Highly recommend submitting 😎

[Christian Dallago @sacdallago]

Nine days left to submit your work to the GenBio workshop for ICML!!!

https://t.co/bMtltd5SN1

[Benjamin Perry @bots_and_bits]

@tw_killian Hey Taylor, will definitely stop by to check it out and connect!

[Benjamin Perry @bots_and_bits]

@omnicui true. I think the assumption is these models are trained on the wet lab data, but this review assesses how they perform after that

[Benjamin Perry @bots_and_bits]

I'll be in Rio this week for #ICLR2026 🇧🇷 including presenting AlphaFast and LIghtning-Boltz at the GEM workshop (4/27 in room 210)

DMs are open to chat about all things accelerating bio-molecular structure prediction and contrastive learning for protein design!🧬

[Benjamin Perry @bots_and_bits]

@anindyadeeps So I believe your concerns about DBs would apply to the upstream Boltz as well. (see scatterplots: we compare against the original MSA web server)

However, if there are specific DBs you care about let me know. That’s definitely something we can expand for pretty easily!

2/2

[Benjamin Perry @bots_and_bits]

Boltz-2 just got a major speed upgrade. 🚀

We’re releasing Lightning-Boltz, a local, GPU-accelerated framework free from public MSA server bottlenecks.⚡

On a single L40S, total runtime drops to 28s per input vs 89s with the rate-limited server and 298s with MMseqs-CPU.

1/5 🧵

[Benjamin Perry @bots_and_bits]

@anindyadeeps This is a great point! The upstream boltz-2 uses the ColabFold/MMseqs2 server (https://t.co/wjvqo6a7Vm).

Our goal here was to port the ColabFold batch-search workflow into Boltz locally with MMseqs2-GPU to do direct apples-to-apples speedup.

1/2

[Benjamin Perry @bots_and_bits]

@anindyadeeps great question! LightningBoltz uses local execution of MMSeqs2-GPU MSA construction rather than the MSA server.

We have both options for using the MSA server and CPU based local search as a backup, but the biggest speed gain is if you have a GPU on your system.

[Benjamin Perry @bots_and_bits]

Wet lab validation is critical for actual success of AI models in biology!🧪

[Clay Kosonocky @kosonocky]

Have you wondered what the wet lab success rates are for current AI-driven protein design models? Look no further!

In our new open access review, @KevinKaichuang, @avapamini, @SarahAlamdari, and I report wet lab success rates for *over 200* different protein design tasks 🧬💻